RDF geeking for Life Sciences

W3C Semantic Web in Health Care and Life Sciences.

Eric Prud'hommeaux, Sanitation Engineer.
URL: http://tinyurl.com/ybdsnme
Last modified: $Date: 2009/10/05 12:22:07 $

This work is licensed under a Creative Commons Attribution 3.0 License, with attribution to W3C.

RDF: fabric of assertions about people, ...

... medicine, ...

... life sciences, ...

... in one universe.

Who is HCLS?

W3C Semantic Web Health Care and Life Sciences Interest Group.

90+ members from industry and academia.

a couple KBs (order 400 triples)

papers

IG notes
BMC Bioinformatics, AMIA, ...

Workshops

Ontology mapping code.

6 Task Forces

BioRDF – integrated neuroscience knowledge base
Linking Open Drug Data – aggregation of Web-based drug data
Clinical Observations Interoperability – patient recruitment
Scientific Discourse – building communities through networking
Terminology – SemWeb representation of existing resources
Translational Medicine Ontology – patient-centric upper level

Other Projects: Clinical Decision Support...

HCLS KB

A Knowledgebase for Neuroscience

Built by Neurocommons and the HCLS BioRDF task force.

Precise answers to complex questions:

What proteins are associated with pyramidal neurons
AND are involved in signal transduction.

Integrate Databases

Mesh
Pubmed
Entrez Gene
Gene Ontology
...

Real Query

Linking Open Drug Data

Clinical Observations Interoperability

Bridge relational databases to HL7/RIM.
Bridge HL7/RIM to SDTM.
Encode some clinical protocol (inclusion/exclusion criteria) in SDTM.
Use drug ontology to map drug classes to brands.
Protocol selection in SDTM translated to RIM, relational query.
Compare this SemWeb-y approach to other approaches.

Real Hospital Database

Observation O
on Patient P
by Doctor D.

Real Hospital Database as RDF

Observation O
on Patient P
by Doctor D.

... Sharable RDF

   PREFIX :mydb <http://cityhospital.example/dbs>
CONSTRUCT { ?o a               study:SubjectObservation .
            ?o study:subject   ?p .
            ?o study:clinician ?d .
            ?d :foaf:name ?dName }

    WHERE { ?o mydb:patient ?p .
            ?o mydb:doctor  ?d .
            ?d mydb:name    ?dName }

Query Transformation

PREFIX db: <…>
SELECT ?o ?d
 WHERE {
   ?o db:onPatient <Bob> .
   ?o db:byDoctor ?d .
 }

PREFIX db: <…>
SELECT observations... doctors...
  FROM observations, 
   patients ON id=observations.patient ...
   WHERE patients.name = "Bob"
 }

Scientific Discourse ...

Scientific Discourse expresses hypotheses

Scientists use this!

Terminology

Importing terms into Semantic Web.

Start with pharmacological classifications in SNOMED-CT and UMLS
Create URL identifiers for terms.
Express term relationships (e.g. pre-coordinated terms) in OWL.
Document process and and expressivity of schema languages.

Translational Medicine Ontology

Requires data from bench to bedside.

vested parties (mostly from pharma)
specific use cases
identify data sources
code as queries
voila, practical ontology

Spans health care and life sciences

Models drug development lifecycle

Unification needed in lots of places

Protein/Gene identifiers
Biological processes
Publication abstracts
Publication metadata
Chemical identifiers
Patient data
...

a course view of DBpedia

a course view of LODD

Inference

SemWeb designed for inference.
Use sparingly
Standards encourage semantic interoperability.

Inferring types

type labels classify things in your data.
- :Fido a :dog .
subtype creates hierarchy of these types.
- :dog rdfs:subClassOf :mammal .
- ⇒ :Fido a :mammal .
subProperty does the same for predicates.
- :Bob my:friendsWith :Sue .
- my:freindsWith rdfs:subProperyOf foaf:knows .
- ⇒ :Bob foaf:knows :Sue .

Inferring types

domain/range identify the type of the subject/object of a predicate.
- _:B123 foaf:givenName "Bob" .
- foaf:givenName rdfs:domain foaf:Person .
- ⇒ _:B123 a foaf:Person .
does not "validate" ...
- :Logan airport:code "KBOS" .
- :Logan foaf:givenName "Logan International Airport" .
... it "learns"
- ⇒ :Logan a foaf:Person .

OWL - inferring lots of stuff

transitiveProperty: if A p1 B and B p1 C, then A p1 C.
- :Bob :worksFor :Sue .
- :Sue :worksFor :Tom .
- ⇒ :Bob :worksFor :Tom .
Derivations of types.
Derivations of identity.

Introduction to OWL

Type inference.
- :Bob a :MGHcharlesStOncologist .
Identity inference.
- :Bob owl:sameAs :BobSmith .
Consitency rules.
- An :MGHcharlstStPatient
  - is an :MGHpatient.
  - has a :physician of type (:MGHcharlesStOncologist or :MGHcharlesStOptician or ...).
  - has an MGH Charles Street consent form.

OWL consistency rules

An :MGHcharlstStPatient
- is an :MGHpatient.
- has one :physician of type (:MGHcharlesStOncologist or :MGHcharlesStOptician or ...).
- has an MGH Charles Street consent form.
Limited "validation" capability.
- _:patientX :physician _:physY
  _:physY a :MGHcharlesStOncologist
- _:patientX :physician _:physZ
  _:physZ a :MGHcharlesStOpticiaMn
- ⇒ :MGHcharlesStOncologist is the same as :MGHcharlesStOptician
use disjointWith to assert that the class of :MGHcharlesStOncologists has different members than the the class :MGHcharlesStOpticians
(use differentFrom to assert that :MGHcharlesStOncologist7 is not the same as :MGHcharlesStOptician3)

OWL type inference

onProperty
- :fluffy :hasChild :scratcher .
- :scratcher a :kitten .
- ⇒ :fluffy a :cat .

OWL identity inference

inverseFunctionalPropery
- :Bob foaf:mbox <mailto:bobS@foo.example> .
- :BobSmith foaf:mbox <mailto:bobS@foo.example> .
- foaf:mbox a owl:inverseFunctionalProperty .
- ⇒ :Bob is the same as :BobSmith
cardinality constraints
- :patientEncounter5 :patient :Bob .
- :patientEncounter5 :patient :BobSmith .
- :patient owl:maxCardinality 1 .
- ⇒ :Bob is the same as :BobSmith

OWL expression

An :MGHcharlstStPatient is an :MGHpatient who has one :physician of type (:MGHcharlesStOncologist or :MGHcharlesStOptician or ...)

Abstract syntax

Class(a:MGHcharlstStPatient partial a:MGHpatient)
Class(a:MGHcharlstStPatient complete 
    restriction(a:physician
        allValuesFrom(unionOf(a:MGHcharlesStOncologist a:MGHcharlesStOptician))))

RDF syntax:

_:MCSO rdfs:subClassOf :MGHpatient .
_:MCSO rdfs:subClassOf _:physType .
_:physType owl:onProperty :physician .
_:physType owl:allValuesFrom _:list1 .
_:list1 rdf:first :MGHcharlesStOncologist .
_:list1 rdf:rest _:list2 .
_:list2 rdf:first :MGHcharlesStOptician .
_:list2 rdf:rest rdf:nil .
_:MCSO owl:equivalentClass :MGHcharlstStPatient .

OWL restrictions

_:physType a owl:Restriction .
_:physType owl:onProperty :physician .
_:physType owl:allValuesFrom :HasScalpel .

OWL restrictions

_:physType a owl:Restriction .
_:physType owl:onProperty :physician .
_:physType owl:allValuesFrom _:anotherRestriction .

OWL restrictions

_:physType a owl:Restriction .
_:physType owl:onProperty :physician .
_:physType owl:allValuesFrom _:anotherRestriction .
_:anotherRestriction a owl:Restriction .
_:anotherRestriction owl:onProperty :has .
_:anotherRestriction owl:allValuesFrom _:KnownScalpels .

RDF Formats

Turtle
RDF/XML
RDFa
GRDDL
JSON
SocialStream
RDBs

Turtle Example: Blood Pressure






<encounter7> edns:patient <patient3> .

<patient3>   r:type galen:Patient ;
             foaf:family_name "Levin" ;
             foaf:firstName "Henry" .

Turtle Example: Blood Pressure






<encounter7> edns:patient <patient3> ;
             edns:screeningBP <s20090714c> .
<patient3>   r:type galen:Patient ;
             foaf:family_name "Levin" ;
             foaf:firstName "Henry" .
<s20090714c> dc:date "2009-07-14T18:23"^^xsd:dateTime ;
             edns:systolic "132"^^edns:mmHg ;
             edns:diastolic "86"^^edns:mmHg .

Turtle Example: Blood Pressure

@prefix edns: <http://www.loa-cnr.it/ontologies/ExtendedDnS.owl#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .
@prefix galen: <http://www.co-ode.org/ontologies/galen#> .
@prefix snomed: <http://termhost.example/SNOMED/> .

<encounter7> edns:patient <patient3> ;
             edns:screeningBP <s20090714c> .
<patient3>   r:type galen:Patient ;
             foaf:family_name "Levin" ;
             foaf:firstName "Henry" .
<s20090714c> dc:date "2009-07-14T18:23"^^xsd:dateTime ;
             edns:systolic "132"^^edns:mmHg ;
             edns:diastolic "86"^^edns:mmHg .

Turtle Example: Blood Pressure

@prefix edns: <http://www.loa-cnr.it/ontologies/ExtendedDnS.owl#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .
@prefix galen: <http://www.co-ode.org/ontologies/galen#> .
@prefix snomed: <http://termhost.example/SNOMED/> .

<encounter7> edns:patient <patient3> ;
             edns:screeningBP <s20090714c> .
<patient3>   r:type galen:Patient ;
             foaf:family_name "Levin" ;
             foaf:firstName "Henry" .
<s20090714c> dc:date "2009-07-14T18:23"^^xsd:dateTime ;
             edns:systolic "132"^^edns:mmHg ;
             edns:diastolic "86"^^edns:mmHg ;
             edns:posture snomed:_163035008 . # SNOMED:sitting

RDF/XML Example: Blood Pressure

<rdf:RDF xmlns:edns="http://www.loa-cnr.it/ontologies/ExtendedDnS.owl#"
	 xmlns:foaf: "http://xmlns.com/foaf/0.1/" .
	 xmlns:galen: "http://www.co-ode.org/ontologies/galen#" .
	 xmlns:snomed: "http://termhost.example/SNOMED/">
<rdf:Description rdf:about="encounter7">
  <edns:patient rdf:resource="patient3"/>
  <edns:screeningBP>s20090714c</edns:screeningBP>
</rdf:Description>
<galen:Patient rdf:about="patient3" foaf:family_name="Levin" foaf:firstName="Henry">
<rdf:Description rdf:about="s20090714c">
  <dc:date xsi:dt="http://www.w3.org/2001/XMLSchemadateTime">2009-07-14T18:23</dc:date>
  <edns:systolic xsi:dt="http://www.loa-cnr.it/ontologies/ExtendedDnS.owl#mmHg">132</edns:systolic>
  <edns:diastolic xsi:dt="http://www.loa-cnr.it/ontologies/ExtendedDnS.owl#mmHg">86</edns:systolic>
  <edns:posture rdf:resource="http://termhost.example/SNOMED/_163035008</edns:posture> <!-- SNOMED:sitting -->
</rdf:Description>

RDFa

Why not merge web of documents and web data?
RDFa (RDF in attributes) allows exactly this.
this document uses RDFa
it hasn't killed anyone yet
RDFa goals:
- modularity
- extensibility
- loose coupling
- don't repeat yourself (DRY)
- in-context metadata (copy & paste)

RDFa Example: Chemicals

InChI is a textual identifier for chemical substances. Consider inchi.html:

<table>
<tr>
  <th>Familiar name</th><th>InChI</th>
</tr><tr>
  <td>Methane</td>
  <td about="http://example.org/methane" property="chem:inchi"
      xmlns:chem="http://www.blueobelisk.org/chemistryblogs/">
    InChI=1/CH4/h1H4
  </td>
...

This RDFa encodes the single RDF triple:

<http://example.org/methane> chem:inchi "InChI=1/CH4/h1H4" .

RDFa: The Gory Details

attribute	specifies	attribute	specifies
`@about`	subjects	`@property`	predicate relating subject to literal content
`@href`	objects, clickable	`@rel`	predicate relating subject to resources (`@href`, `@src`)
`@src`	objects, embedded	`@rev`	predicate relating resources to subject in reverse
`@resource`	objects, not clickable	`@content`	Object of triple (instead of element content)
`@instanceof`	RDF types	`@datatype`	literal values' data types

For more, see the RDFa primer or the RDFa specification.

Using RDFa: In context

See inchi.html.

Introduction to GRDDL

Using GRDDL to get RDF from XML, XHTML
XML data
Mapped to RDF
XSLT
GRDDLing HTML
GRDDL Mechanics

Using GRDDL to get RDF from XML, XHTML

GRDDL (Gleaning Resource Descriptions from Dialects of Languages)
boostrap RDF out of XML (e.g. XHMTL)
Source Document: references at least one GRDDL transformation.
GRDDL Transformation: an (XSLT) transformation program.

XML data

<?xml version="1.0"?>
<ClinicalDocument transformation="hl7-rim-to-pomr.xslt">
  <recordTarget>
    <patientRole>
      <patientPatient>
	<name>
	  <given>Henry</given>
	  <family>Levin</family>
	</name>
	<administrativeGenderCode code="M"/>
	<birthTime value="19320924"/>
      </patientPatient>
    </patientRole>
  </recordTarget>
  <component>
    <StructuredBody>
      <Observation>
	<code displayName="Cuff blood pressure"/>
	<effectiveTime value="200004071430"/>
	<targetSiteCode displayName="Left arm"/>
	<entryRelationship typeCode="COMP">
	  <Observation>
	    <effectiveTime value="200004071530"/>
	    <value value="132" unit="mm[Hg]"/>
	  </Observation>
	</entryRelationship>
      </Observation>
      <Observation>
	<code displayName="Cuff blood pressure"/>
	<effectiveTime value="200004071530"/>
	<targetSiteCode displayName="Left arm"/>
	<entryRelationship typeCode="COMP">
	  <Observation>
	    <code displayName="Systolic BP"/>
	    <effectiveTime value="200004071530"/>
	    <value value="135" unit="mm[Hg]"/>
	  </Observation>
	</entryRelationship>
	<entryRelationship typeCode="COMP">
	  <Observation>
	    <code displayName="Diastolic BP"/>
	    <effectiveTime value="200004071530"/>
	    <value value="88" unit="mm[Hg]"/>
	  </Observation>
	</entryRelationship>
      </Observation>
    </StructuredBody>
  </component>
</ClinicalDocument>

Patient history
- Name
- Age
- Illnesses
Encounter details
- Reason for visit
- Vitals (e.g. BP)
- Tests
- Prescriptions
Facility details...

Mapped to RDF

XSLT

<xsl:template match="rim:ClinicalDocument[rim:recordTarget/
                        rim:patientRole/rim:patientPatient]">
  <cpr:patient-record>
    <xsl:apply-templates select="rim:effectiveTime"/>
    <xsl:apply-templates select="rim:recordTarget/
                        rim:patientRole/rim:patientPatient"/>
    <xsl:for-each select="rim:author/
                      rim:assignedAuthor/rim:assignedPerson">
      <foaf:maker>
        <foaf:Person>
          <xsl:apply-templates select="rim:name"/>
        </foaf:Person>
      </foaf:maker>
    </xsl:for-each>
    <xsl:apply-templates select="rim:component"/>
  </cpr:patient-record>
</xsl:template>

<xsl:template match="rim:name/rim:family">
  <foaf:family_name><xsl:value-of select="."/></foaf:family_name>
</xsl:template>

<xsl:template match="rim:name/rim:given">
  <foaf:firstName><xsl:value-of select="."/></foaf:firstName>
</xsl:template>

<xsl:template match="rim:patientPatient">
  <edns:about>
    <galen:Patient>
      <xsl:apply-templates select="rim:name"/>
    </galen:Patient>
  </edns:about>
</xsl:template>

Patterns for extracting information from XML.
XML or plain text output
Templates matched by XPath
- Most specific rule applies.
Explicit enumerators
- xsl:for-each, xsl:when...

XSLT

<xsl:template match="rim:ClinicalDocument[rim:recordTarget/
                        rim:patientRole/rim:patientPatient]">
  <cpr:patient-record>
    <xsl:apply-templates select="rim:effectiveTime"/>
    <xsl:apply-templates select="rim:recordTarget/
                        rim:patientRole/rim:patientPatient"/>
    <xsl:for-each select="rim:author/
                      rim:assignedAuthor/rim:assignedPerson">
      <foaf:maker>
        <foaf:Person>
          <xsl:apply-templates select="rim:name"/>
        </foaf:Person>
      </foaf:maker>
    </xsl:for-each>
    <xsl:apply-templates select="rim:component"/>
  </cpr:patient-record>
</xsl:template>

<xsl:template match="rim:name/rim:family">
  <foaf:family_name><xsl:value-of select="."/></foaf:family_name>
</xsl:template>

<xsl:template match="rim:name/rim:given">
  <foaf:firstName><xsl:value-of select="."/></foaf:firstName>
</xsl:template>

<xsl:template match="rim:patientPatient">
  <edns:about>
    <galen:Patient>
      <xsl:apply-templates select="rim:name"/>
    </galen:Patient>
  </edns:about>
</xsl:template>

XSLT

<xsl:template match="rim:ClinicalDocument[rim:recordTarget/
                        rim:patientRole/rim:patientPatient]">
  <cpr:patient-record>
    <xsl:apply-templates select="rim:effectiveTime"/>
    <xsl:apply-templates select="rim:recordTarget/
                        rim:patientRole/rim:patientPatient"/>
    <xsl:for-each select="rim:author/
                      rim:assignedAuthor/rim:assignedPerson">
      <foaf:maker>
        <foaf:Person>
          <xsl:apply-templates select="rim:name"/>
        </foaf:Person>
      </foaf:maker>
    </xsl:for-each>
    <xsl:apply-templates select="rim:component"/>
  </cpr:patient-record>
</xsl:template>

<xsl:template match="rim:name/rim:family">
  <foaf:family_name><xsl:value-of select="."/></foaf:family_name>
</xsl:template>

<xsl:template match="rim:name/rim:given">
  <foaf:firstName><xsl:value-of select="."/></foaf:firstName>
</xsl:template>

<xsl:template match="rim:patientPatient">
  <edns:about>
    <galen:Patient>
      <xsl:apply-templates select="rim:name"/>
    </galen:Patient>
  </edns:about>
</xsl:template>

XSLT

<xsl:template match="rim:ClinicalDocument[rim:recordTarget/
                        rim:patientRole/rim:patientPatient]">
  <cpr:patient-record>
    <xsl:apply-templates select="rim:effectiveTime"/>
    <xsl:apply-templates select="rim:recordTarget/
                        rim:patientRole/rim:patientPatient"/>
    <xsl:for-each select="rim:author/
                      rim:assignedAuthor/rim:assignedPerson">
      <foaf:maker>
        <foaf:Person>
          <xsl:apply-templates select="rim:name"/>
        </foaf:Person>
      </foaf:maker>
    </xsl:for-each>
    <xsl:apply-templates select="rim:component"/>
  </cpr:patient-record>
</xsl:template>

<xsl:template match="rim:name/rim:family">
  <foaf:family_name><xsl:value-of select="."/></foaf:family_name>
</xsl:template>

<xsl:template match="rim:name/rim:given">
  <foaf:firstName><xsl:value-of select="."/></foaf:firstName>
</xsl:template>

<xsl:template match="rim:patientPatient">
  <edns:about>
    <galen:Patient>
      <xsl:apply-templates select="rim:name"/>
    </galen:Patient>
  </edns:about>
</xsl:template>

XSLT courtesy of Chimezie Ogbuji

GRDDLing HTML

GRDDL can extract RDF from both XML and (X)HTML.

<Observation>
  <effectiveTime value="20090403T12:34+0400">
    <title>Clinical Study 8B1a: Patient BP</title>
    <value value=86 unites="mm[Hg]"/>
  </effectiveTime>
</Observation>

<html>
  <head profile="http://www.w3.org/2003/g/data-view">
    <title>Clinical Study 8B1a: Patient BP</title>
    <link rel="transformation" href="bp-html-to-pomr.xslt" />
  </head>
  ...

GRDDL Mechanics

Content publisher provides XML or XHTML document that does one of:

For individual documents:
- (XHTML) Uses the GRDDL HTML metadata profile and has a link element pointing to a transofmation
- (XML) Has a grddl:transformation attribute on the root node
For families of documents:
- (XHTML) Uses an HTML metadata profile that itself resolves via GRDDL to RDF that points to a profile transformation
- (XML) Has a namespace document that itself resolves via GRDDL to RDF that points to a namespace transformation

GRDDL Resources

Even relational data ...

... produces graphs.

Advanced SPARQL

GRAPH constraints - good for provenance
Result forms - gives you RDF closure
Examples

GRAPH constraints

SPARQL lets us query different RDF graphs in a single query. Consider movie reviews:

Target one authoritative data source (What does Roger Ebert say?):

GRAPH <http://example.org/reviews/rogerebert> {
  ex:atonement rev:hasReview ?review .
  ?review rev:rating ?rating .
}

Relate multiple sources (How do my reviews compare to Ebert's?):

GRAPH <http://example.org/reviews/rogerebert> {
  ?movie rev:hasReview ?rev1 .
  ?rev1 rev:rating ?ebert .
}
GRAPH <http://example.org/reviews/me> {
  ?movie rev:hasReview ?rev2 .
  ?rev2 rev:rating ?me .
}

Retrieve provenance data (Which reviewers have given out perfect ratings?):
- ```
GRAPH ?reviewer_graph {
  ?review rev:rating 10 .
}
```

Result forms

3 types of queries:

ASK -> boolean: does the query have any results?
CONSTRUCT -> query-templated RDF graph: uses variable bindings to return new RDF triples
DESCRIBE - server's choice RDF graph

SELECT and ASK results can be returned as XML or JSON. CONSTRUCT and DESCRIBE results can be returned via any RDF serialization (e.g. RDF/XML or Turtle).

SPARQL Protocol Mechanics

The SPARQL Protocol is a simple method for asking and answering SPARQL queries over HTTP. A SPARQL URL is built from three parts:

The URL of a SPARQL endpoint (e.g. http://example.org/sparql)
(Optional, as part of the query string) The graphs to be queried against (e.g. named-graph-uri=http://example.org/reviews/ebert
(As part of the query string) The query itself (e.g. query=SELECT...)

http://example.org/sparql?named-graph-uri=http%3A%2F%2Fexample.orgm%2F
reviews%2Febert&query=SELECT+%3Freview_graph+WHERE+%7B%0D%0A++GRAPH+%3Frev
iew_graph+%7B%0D%0A+++++%3Freview+rev%3Arating+10+.%0D%0A++%7D%0D%0A%7D

SPARQL Resources

SPARQL Frequently Asked Questions
SPARQL implementations - community maintained list of open-source and commercial SPARQL engines
Public SPARQL endpoints - community maintained list
SPARQL extensions - collection of SPARQL extensions implemented in various SPARQL engines

Internationalization (i18n)

Unicode is a big alphabet.
Node identifiers are IRIs.
RDF/XML leans on XML declaration.
- <?xml version="1.0" encoding="UTF-8"?>
SPARQL is always UTF-8.

Collaboration

We want your help

excited group of people
real-world impact
fun scaling problems
lots of papers
good post-grad contacts

join per <http://www.w3.org/2001/sw/hcls/#Participation>.